Cursor 2.0 (cursor.com)

Cursor 2.0 is a major update focused on scalable, safer, and faster agentic coding workflows. It adds an agent/plan sidebar, the ability to run up to eight agents in parallel on a single prompt (each in an isolated git worktree or remote copy of your repo to avoid file conflicts), and a new first-party agentic coding model, Composer, which the company claims is ~4x faster than similarly capable models. The browser agent is now generally available and can be embedded in-editor with tools to select DOM elements and forward that info to agents. Team management features include centrally defined commands/rules, sharable deeplinks for prompts and rules, voice control (speech-to-text and custom submit keywords), and timestamped admin logs for governance. Technical improvements emphasize performance, safety, and enterprise controls: sandboxed terminals are GA on macOS and run by default for non-allowlisted shell commands (workspace read/write but no network), cloud agents now promise 99.9% reliability and instant startup, and the agent harness has been overhauled with notable quality gains (especially for GPT-5 Codex). Language Server Protocols load and run much faster—Python and TypeScript LSPs now dynamically increase memory limits for large projects—and memory leaks were fixed. Other productivity upgrades include easier multi-file diffs, inline file pills, fewer manual context attachments (agents can self-gather context), and flexible plan/build workflows (different models for planning vs building, foreground/background builds, parallel plans). Together, these changes aim to make multi-agent development fast, reproducible, auditable, and safer for teams building with AI.